WEBVTT Kind: captions Language: en 00:00:00.140 --> 00:00:00.820 AI 00:00:01.060 --> 00:00:05.040 The concept of AI, it's something that I think we're all familiar with 00:00:05.580 --> 00:00:10.520 Thanks to numerous amount of TV shows, movies and novels 00:00:10.740 --> 00:00:13.680 For me, personally I find this to be sort of a tired 00:00:13.720 --> 00:00:17.820 concept at this point. You've just seen it so many times in sci-fi. 00:00:17.920 --> 00:00:22.000 And especially the idea that AI is going to take over. 00:00:22.000 --> 00:00:24.460 We've seen in it "Terminator" 00:00:31.200 --> 00:00:35.980 But I've also seen this in some of my favorite work of fiction like "A Space Odyssey" 00:00:35.980 --> 00:00:37.980 "Open the pod bay doors, HAL." 00:00:38.680 --> 00:00:42.180 "I'm sorry Dave. I'm afraid I can't do that." 00:00:42.800 --> 00:00:47.840 Some of my favorite novels like, "I Have No Mouth And I Must Scream". Don't get me wrong. 00:00:47.840 --> 00:00:51.220 I find it a very interesting concept, even though it's overused, but 00:00:51.220 --> 00:00:54.500 when Stephen Hawking, in 2014 came out and said, 00:00:54.520 --> 00:01:01.140 "But I think the development of full artificial intelligence could spell the end of the human race," 00:01:01.360 --> 00:01:07.080 My mind sort of went "Pfft, Stephen Hawkins, what do you know about anything? Okay?" 00:01:07.440 --> 00:01:10.400 "Do you watch Rick and Morty? Because, I do, 00:01:10.400 --> 00:01:15.100 and I think I have a little better grasp of the universe concept ideas, like AI. 00:01:15.340 --> 00:01:16.860 Thank you very much, Stephen. 00:01:16.940 --> 00:01:19.740 But, uh, stick to your science stuff, all right?" 00:01:19.740 --> 00:01:25.740 The concept of AI taking over... It feels so far off, because it seems so... obscure. 00:01:25.820 --> 00:01:29.060 The way, it's portrayed in Hollywood movies or in the work of fiction. 00:01:29.380 --> 00:01:30.980 But in reality, it's actually... 00:01:31.080 --> 00:01:35.120 I think, or from what I've learned, an actual threat. 00:01:35.260 --> 00:01:41.140 The idea that Johnny Depp could come back and kill us all, is not as far away as we think. 00:01:42.620 --> 00:01:51.660 *scenes from the movie "Transcendence"* 00:01:52.520 --> 00:01:53.920 But how did we come to this conclusion? 00:01:53.960 --> 00:01:56.300 I'm going to try to explain, okay? 00:01:56.400 --> 00:02:00.260 But I truly don't know shit what I'm talking about, so please... 00:02:00.500 --> 00:02:04.600 If we take it back a couple steps, a lot of couple steps, uh... 00:02:04.620 --> 00:02:08.620 There's this game of tic-tac-toe, that I found. 00:02:08.620 --> 00:02:16.920 That, no matter what input you make, the computer will never let you win. 00:02:17.040 --> 00:02:18.420 It's kind of fucking annoying. 00:02:18.800 --> 00:02:20.440 It's programmed with algorithms, 00:02:20.520 --> 00:02:23.120 So that, no matter what move I make, 00:02:23.200 --> 00:02:29.080 it knows exactly what move to counter it to make sure that I can't win, no matter what. 00:02:29.080 --> 00:02:32.300 Not a big deal, not a big deal. Okay? 00:02:32.300 --> 00:02:37.460 In 1958, A.H. Simon and Allen Newell, AI experts... 00:02:37.500 --> 00:02:40.940 But, what if you take a more complex game than tic-tac-toe? 00:02:40.940 --> 00:02:42.560 Say... chess for example? 00:02:42.560 --> 00:02:45.840 There's a lot more possible outcomes in that game. 00:02:45.880 --> 00:02:47.520 Foresaw, that within ten years, 00:02:47.520 --> 00:02:51.080 a digital computer will beat the world's best chess champion. 00:02:51.080 --> 00:02:54.460 Now, it didn't take ten years. Not until 1997. 00:02:54.460 --> 00:02:55.480 You may have heard of this. 00:02:55.480 --> 00:02:58.440 It was quite the big deal at the time. 00:02:58.440 --> 00:03:04.660 Deep Blue became the first computer that was able to beat the reigning chess champion at the time, Kasparov. 00:03:04.820 --> 00:03:10.660 "But it's clear that the computer will reliably do what he himself would do, 00:03:10.660 --> 00:03:13.960 and he recognizes, that he has already lost. 00:03:13.960 --> 00:03:19.980 On Deep Blue's 19th move, the champion resigns." 00:03:20.620 --> 00:03:22.800 Now, it still doesn't seem like that big of a deal, 00:03:22.800 --> 00:03:24.900 and basically the way Deep Blue worked was, 00:03:24.900 --> 00:03:25.980 that it would scan 00:03:25.980 --> 00:03:28.900 every single possible outcome it could make, 00:03:28.900 --> 00:03:31.480 about 200,000 per second. 00:03:31.480 --> 00:03:33.360 And it would make the best decision 00:03:33.360 --> 00:03:37.700 based on what he could find through this method of scanning. 00:03:37.700 --> 00:03:40.700 At this point, I'm still like Stephen Hawking ... 00:03:40.700 --> 00:03:44.360 I've seen the videos of the machines falling over, okay? 00:03:44.380 --> 00:03:46.580 I think we have nothing to worry about. 00:03:46.580 --> 00:03:49.260 But here's where I think it gets interesting. 00:03:49.520 --> 00:03:51.520 In March 15, 2016, 00:03:51.640 --> 00:03:54.960 The champion of the Chinese board game "go", 00:03:54.960 --> 00:03:56.640 was beaten by an AI. 00:03:56.720 --> 00:03:58.080 "Against AlphaGo, 00:03:58.080 --> 00:04:01.240 the artificial intelligence, designed by Google's DeepMind. 00:04:01.780 --> 00:04:04.840 It was a resounding loss. They had won only one game. 00:04:04.840 --> 00:04:11.280 'AlphaGo wins! We landed it on the moon. So proud of the team! Respect to the amazing Lee Sedol too.' " 00:04:11.280 --> 00:04:13.740 Now, the reason why this is such a big deal is that 00:04:13.740 --> 00:04:15.820 in chess, you only have so many options. 00:04:15.820 --> 00:04:17.180 But in go, 00:04:17.180 --> 00:04:20.890 there are so many different moves that you can make, 00:04:20.960 --> 00:04:23.820 there are more possible moves that you can make than there are atoms in the universe 00:04:23.820 --> 00:04:24.900 and there's just no way 00:04:24.900 --> 00:04:29.240 that you're going to be able to compute that amount of options, 00:04:29.240 --> 00:04:31.700 to figure out what's the best move to make. 00:04:31.700 --> 00:04:32.860 So how did they make this? 00:04:32.860 --> 00:04:35.540 It may not seem like that big of a deal, either? 00:04:35.540 --> 00:04:37.820 But it's really cool. Okay? It's really cool. 00:04:37.820 --> 00:04:40.480 It basically uses deep reinforcement learning. 00:04:40.480 --> 00:04:43.500 Which is similar to how we learn as humans. 00:04:43.840 --> 00:04:45.420 Through trial and error, 00:04:45.420 --> 00:04:47.000 reward and punishment, 00:04:47.000 --> 00:04:48.660 and raw inputs. 00:04:48.660 --> 00:04:50.940 Say, if we see something ourselves, 00:04:51.260 --> 00:04:54.280 the computer figure learns itself, 00:04:54.280 --> 00:04:55.880 how to become good at the game. 00:04:55.880 --> 00:04:56.800 Not too long ago, 00:04:56.800 --> 00:04:58.520 there was a viral video of- 00:04:58.520 --> 00:05:01.060 From SethBling, that uses method 00:05:01.060 --> 00:05:03.900 to teach a computer to play Mario. 00:05:03.900 --> 00:05:06.180 And it became really fucking good at it. 00:05:07.120 --> 00:05:09.120 REALLY good at it. 00:05:09.860 --> 00:05:11.450 Look at that. 00:05:11.450 --> 00:05:13.840 Basically, it uses neural networks 00:05:13.840 --> 00:05:15.840 to learn how to play the game. 00:05:16.460 --> 00:05:19.380 Which is similar to how we think as human beings 00:05:19.380 --> 00:05:21.400 And with enough computing power, 00:05:21.400 --> 00:05:24.760 you could simulate a human brain, in this way. 00:05:24.760 --> 00:05:26.140 But we're not there yet. 00:05:26.160 --> 00:05:28.140 But it wasn't good from the beginning, 00:05:28.140 --> 00:05:29.760 it had to learn how to get good. 00:05:29.760 --> 00:05:30.900 GIT GUD. 00:05:30.900 --> 00:05:34.260 In the beginning, it doesn't even know where it has to go 00:05:34.380 --> 00:05:35.480 or what the option is 00:05:35.480 --> 00:05:36.480 or what Mario is. 00:05:36.620 --> 00:05:39.160 But eventually, it figures out it needs to move right, 00:05:39.160 --> 00:05:42.460 but through different generations and learning 00:05:42.460 --> 00:05:43.920 and from trial and error 00:05:43.920 --> 00:05:45.860 and adapting from these mistakes, 00:05:45.920 --> 00:05:48.880 It eventually, becomes better and better. 00:05:48.880 --> 00:05:51.700 And the similar method was used for the AlphaGo. 00:05:52.060 --> 00:05:54.600 Program where it would train against itself. 00:05:54.600 --> 00:05:57.360 Slowly becoming better and better and better 00:05:57.360 --> 00:05:59.100 and eventually a master at the game. 00:05:59.160 --> 00:06:00.580 There's a super cool video, 00:06:00.580 --> 00:06:03.260 about a robot that doesn't know 00:06:04.000 --> 00:06:05.040 that it has limbs 00:06:05.040 --> 00:06:08.860 But it teaches itself how to walk, despite of this. 00:06:08.860 --> 00:06:11.100 So it's just doing random movements. 00:06:11.240 --> 00:06:13.720 It sort of figures out it has four limbs. 00:06:13.760 --> 00:06:17.940 But it doesn't know where those limbs on its body is attached. 00:06:18.020 --> 00:06:20.060 And by trial and error, 00:06:20.060 --> 00:06:23.180 it eventually figures out where its limbs are positioned. 00:06:23.180 --> 00:06:24.360 And eventually, 00:06:24.360 --> 00:06:27.660 it can very graciously move across. 00:06:27.760 --> 00:06:28.680 That's cool. 00:06:30.020 --> 00:06:32.700 Self learning AI is really fucking cool. 00:06:32.700 --> 00:06:35.440 And there's a lot of advantages that you can do from this. 00:06:35.440 --> 00:06:37.200 Using it in design, for example. 00:06:37.200 --> 00:06:39.820 "This is a 3D printed cabin partition 00:06:39.820 --> 00:06:41.820 that's been designed by a computer. 00:06:42.220 --> 00:06:45.260 It's stronger than the original yet half the weight. 00:06:45.260 --> 00:06:48.400 And it'll be flying in the Airbus A320, later this year. 00:06:48.920 --> 00:06:52.680 So, computers can now generate, they can come up with their own solutions 00:06:52.680 --> 00:06:55.200 to our well-defined problems." 00:06:55.260 --> 00:06:59.000 So then, with Elon Musk as well as Stephen Hawking saying 00:06:59.000 --> 00:07:02.440 AI could become a problem in the future, 00:07:02.440 --> 00:07:05.760 that idea starts to sort of make more sense to me 00:07:05.760 --> 00:07:07.480 knowing this is how it works. 00:07:07.480 --> 00:07:10.220 "I think we should be very careful about artificial intelligence. 00:07:10.220 --> 00:07:15.160 If I would guess at what our biggest existential threat is, 00:07:15.880 --> 00:07:17.880 it's probably that." 00:07:17.920 --> 00:07:20.240 Elon Musk as well as Bill Gates 00:07:20.340 --> 00:07:22.780 chiming in as well, with the same idea. 00:07:22.980 --> 00:07:25.740 But the basic point that Elon has made... 00:07:25.800 --> 00:07:31.380 we have a general purpose learning the algorithm that evolution has endowed us with. 00:07:31.460 --> 00:07:36.140 And it's running in an extremely slow computer. 00:07:36.380 --> 00:07:38.380 Very limited memory size, 00:07:38.380 --> 00:07:43.960 ability to send data to other computers we have to use this funny mouth thing here... 00:07:45.660 --> 00:07:51.340 ...whenever we build a new one it starts over, it doesn't know how to walk. 00:07:53.040 --> 00:07:55.580 So believe me, as soon as this algorithm- 00:07:57.180 --> 00:07:59.760 taking experience and turning it into knowledge- 00:07:59.900 --> 00:08:04.200 Which is so amazing and which we have not done in software. 00:08:06.200 --> 00:08:10.680 As soon as you do that, it's not clear you'll even know when you're just at the human level. 00:08:10.680 --> 00:08:18.400 You'll be at the superhuman level almost as soon as that algorithm is implanted, in silicon. 00:08:18.400 --> 00:08:23.840 Bill basically here, compares how our brains as a computer, 00:08:24.540 --> 00:08:28.860 our method of evolving is very inefficient with- 00:08:29.340 --> 00:08:31.980 comparing it to how AI would be evolving 00:08:31.980 --> 00:08:33.660 and exponentially growing. 00:08:33.660 --> 00:08:36.100 And knowing, keeping that in mind, 00:08:36.100 --> 00:08:37.520 humans are inferior. 00:08:37.520 --> 00:08:39.520 Without a doubt. 00:08:39.640 --> 00:08:40.600 That being said, 00:08:40.600 --> 00:08:43.960 not everyone is on board with this idea that AI is going to take over 00:08:43.960 --> 00:08:46.260 or that it's a problem for the future. 00:08:46.260 --> 00:08:50.220 "What are your thoughts on AI and how it could affect the world? 00:08:51.940 --> 00:08:55.220 You know, I have pretty strong opinions on this. 00:08:55.220 --> 00:08:58.380 I'm really optimistic. Well, I'm an optimistic person in general. 00:08:58.390 --> 00:09:01.659 I think you can build things and the world gets better, but 00:09:02.320 --> 00:09:05.560 with AI especially, I'm really optimistic. 00:09:05.560 --> 00:09:08.420 And I think that people who are naysayers 00:09:08.420 --> 00:09:13.480 and kind of try to drum up these doomsday scenarios are... 00:09:14.440 --> 00:09:18.220 I just I don't understand it. I think it's really negative and 00:09:18.220 --> 00:09:20.880 in some ways, I actually think it's pretty irresponsible." 00:09:22.100 --> 00:09:23.880 Elon Musk responding on Twitter: 00:09:23.880 --> 00:09:28.180 "I've talked to Mark about this. His understanding of the subject is limited." 00:09:29.740 --> 00:09:31.700 I love Mark Zuckerberg, obviously. 00:09:31.800 --> 00:09:33.600 I love the 'zucc' more than anyone. 00:09:33.640 --> 00:09:35.860 It's kind of hard to take him seriously on the subject. 00:09:35.880 --> 00:09:41.020 Especially, since he clearly is trying to make an AI himself. 00:09:41.240 --> 00:09:45.340 "Good morning Mark. It's Saturday. So you only have five meetings. 00:09:45.860 --> 00:09:49.880 Room temperature is set to a cool, 68 degrees." 00:09:49.880 --> 00:09:51.720 I guess what Mark is saying is, 00:09:51.720 --> 00:09:55.640 AI can do a lot for us as humans, it can benefit us greatly. 00:09:55.640 --> 00:09:57.960 And I think what Elon points out is that, 00:09:58.080 --> 00:10:00.940 there are dangers involved with the development on this 00:10:00.940 --> 00:10:02.200 and we need to be careful. 00:10:02.260 --> 00:10:05.140 How can we protect ourselves from ourselves? 00:10:05.400 --> 00:10:07.140 "We are an intelligent adversary. 00:10:07.140 --> 00:10:09.440 We can anticipate threats and plan around them. 00:10:09.440 --> 00:10:12.100 But, so could an super intelligent agent. 00:10:12.100 --> 00:10:13.900 How confident could it be that 00:10:13.900 --> 00:10:15.500 the AI couldn't find a bug. 00:10:15.500 --> 00:10:18.400 Like, given that merely human hackers find bugs all the time. 00:10:18.400 --> 00:10:21.080 I'd say, probably not very confident. 00:10:21.080 --> 00:10:24.080 Like, disconnect the Internet cable to create an air gap. 00:10:24.760 --> 00:10:28.720 But again, like merely human hackers routinely transgress air gaps, 00:10:28.720 --> 00:10:29.800 using social engineering. 00:10:29.800 --> 00:10:31.200 Like right now, as I speak, 00:10:31.200 --> 00:10:33.420 I'm sure there is some employee out there somewhere, 00:10:33.420 --> 00:10:36.160 who is being talked into handing out her account details 00:10:36.160 --> 00:10:38.500 by somebody claiming to be from the IT Department. 00:10:38.500 --> 00:10:40.340 We should not be confident in our ability 00:10:40.340 --> 00:10:43.860 to keep a super intelligent genie locked up in its bottle forever. 00:10:43.960 --> 00:10:46.380 I'm actually fairly optimistic that this problem can be solved. 00:10:46.380 --> 00:10:47.660 Like, we wouldn't have to try 00:10:47.660 --> 00:10:50.400 to write down the long list of everything we care about. 00:10:50.400 --> 00:10:51.360 Or worse yet, 00:10:51.960 --> 00:10:55.720 spell it out in some computer language, like C++ or Python like that. 00:10:55.720 --> 00:10:57.840 That would be a task beyond hopeless. 00:10:57.840 --> 00:11:00.760 Instead, we would create an AI, 00:11:00.760 --> 00:11:04.169 that uses these intelligence to learn what we value. 00:11:05.020 --> 00:11:07.780 And his motivation system is constructed in such a way, 00:11:08.340 --> 00:11:11.660 that it is motivated to pursue our values or, 00:11:11.660 --> 00:11:15.120 to perform actions that it predicts that we would have approved of." 00:11:15.120 --> 00:11:18.640 Computers smarter than human beings is inevitable, 00:11:18.640 --> 00:11:22.600 if you keep in mind how short we have even had technology 00:11:22.600 --> 00:11:24.960 and our presence in the universe. 00:11:24.960 --> 00:11:27.760 Now, whether AI will be something good 00:11:27.760 --> 00:11:31.040 or destroy us all in the future, 00:11:31.060 --> 00:11:33.240 that's just for us to find out. 00:11:33.240 --> 00:11:35.560 Meanwhile, you're going to have to excuse me 00:11:35.560 --> 00:11:38.340 because I have some Rick and Morty episodes to catch up on. 00:11:38.340 --> 00:11:40.900 I hope this video was educational. 00:11:40.900 --> 00:11:43.520 And I hope, I didn't say any wrong things. 00:11:43.520 --> 00:11:45.420 Because I sure am no expert! 00:11:45.420 --> 00:11:48.360 Thank you for leaving a like on this video, if you enjoyed. I really appreciate it. 00:11:48.360 --> 00:11:50.460 Make sure to 'zuccscribe'. 00:11:50.460 --> 00:11:53.660 And as always, squad fam out.